Upload
itam
View
0
Download
0
Embed Size (px)
Citation preview
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
rspb.royalsocietypublishing.org
ResearchCite this article: Esponda F, Gordon DM.
2015 Distributed nestmate recognition in ants.
Proc. R. Soc. B 282: 20142838.
http://dx.doi.org/10.1098/rspb.2014.2838
Received: 18 November 2014
Accepted: 5 March 2015
Subject Areas:behaviour, theoretical biology
Keywords:nestmate recognition, natural computing,
distributed systems, cuticular hydrocarbons
Author for correspondence:Deborah M. Gordon
e-mail: [email protected]
Electronic supplementary material is available
at http://dx.doi.org/10.1098/rspb.2014.2838 or
via http://rspb.royalsocietypublishing.org.
& 2015 The Authors. Published by the Royal Society under the terms of the Creative Commons AttributionLicense http://creativecommons.org/licenses/by/4.0/, which permits unrestricted use, provided the originalauthor and source are credited.Distributed nestmate recognition in ants
Fernando Esponda1 and Deborah M. Gordon2
1Department of Computer Science, Instituto Tecnologico Autonomo de Mexico, Mexico D.F. 01080, Mexico2Department of Biology, Stanford University, CA 94305, USA
We propose a distributed model of nestmate recognition, analogous to the
one used by the vertebrate immune system, in which colony response results
from the diverse reactions of many ants. The model describes how individ-
ual behaviour produces colony response to non-nestmates. No single ant
knows the odour identity of the colony. Instead, colony identity is defined
collectively by all the ants in the colony. Each ant responds to the odour
of other ants by reference to its own unique decision boundary, which is a
result of its experience of encounters with other ants. Each ant thus recog-
nizes a particular set of chemical profiles as being those of non-nestmates.
This model predicts, as experimental results have shown, that the outcome
of behavioural assays is likely to be variable, that it depends on the
number of ants tested, that response to non-nestmates changes over time
and that it changes in response to the experience of individual ants.
A distributed system allows a colony to identify non-nestmates without
requiring that all individuals have the same complete information and
helps to facilitate the tracking of changes in cuticular hydrocarbon profiles,
because only a subset of ants must respond to provide an adequate response.
1. IntroductionDistributed processes are widespread in nature as well as in engineered data
networks [1]. In systems without central control, the role of an individual
part depends on its interactions with other individuals. For example, in the
developing fly brain, lateral inhibition among neighbouring cells regulates
which cells eventually differentiate to become the sensory bristles [2]. Another
example is the vertebrate immune system, in which there is no single cell that is
able to identify every possible antigen; instead individual cells can each detect a
small subset [3]. Because an antigen is encountered by a variety of immune cells
when it enters the organism, there is a high probability that it will be identified,
thus conferring thorough coverage by the aggregate. The immune system can
track changes in ‘self’ efficiently because only a few immune cells must be
updated for every change. This provides a robust security system, difficult to
subvert since there is no single cell where all the information about ‘self’ is
kept and from which it can be stolen, hijacked or copied [4]. Methods for
data privacy and computer security [5–7], like the immune system [8], rely
on fragmenting and distributing sensitive information among a set of agents
so that each one holds only a piece of the puzzle. Their collective characteristics
confer adequate coverage for the host organism; together they define what is
acceptable by individually specifying what is not.
Distributed processes regulate not only the collective behaviour of cells, but
also that of groups of organisms [8]. Here, we propose a model of nestmate
recognition in social insects in which recognition depends on a distributed,
colony-wide process. Nestmate recognition has been observed in many species
of social insects [9], including wasps [10,11] and bees [12–15]. Our model
focuses on ants but may be broadly applicable to other taxa. Ants are an extre-
mely diverse and widespread taxon of more than 14 000 species. All ant species
live in colonies, and nestmate recognition is crucial in regulating colony cohe-
sion and interactions with other colonies. Ants are ecologically important in
every terrestrial ecosystem, and competition among colonies has important
0
0.2
0.4
0.6
0.8
1.0
0.2 0.4 0.6chemical 1
chem
ical
2
0.8 1.0
chemical profile 2
chemical profile 3
chemical profile 1
Figure 1. CHC-space for two chemicals. Each point represents a particularcombination of the chemicals 1 and 2.
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
2
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
effects on ant population dynamics (e.g. [16]). In many
species, colonies engage in ongoing interactions with neigh-
bouring colonies that are crucial for the partitioning of
resources [17–19].
Many studies of nestmate recognition show considerable
variation among individuals in the extent of aggression (e.g.
[13,20,21]). Studies of nestmate recognition in ants rely on be-
havioural assays, with individuals from different colonies
placed in proximity, to determine when one ant identifies
another as belonging to a different colony. Recognition is con-
sidered to occur if individuals are more aggressive towards
non-nestmates than towards nestmates. The results of such
studies are notoriously variable (e.g. [20,22–24]). The out-
comes depend on whether the experiments were conducted
blind [24], on the worker task [23,25] and on the numbers of
ants included in the assay [20]. Often individuals of the same
colony differ in response (e.g. [26]).
The variability among individuals in their aggressive
response to individuals from other colonies has been treated
as a methodological problem rather than as a fundamental
feature of the process of nestmate recognition. The variability
is probably owing to many factors, such as variation among
individuals in their ability to detect chemical profiles [27]
and in differences among task groups in the likelihood that
they will respond in a bioassay [23]. However, current under-
standing of nestmate recognition is built on the premise that
variation among individuals of the same colony in response
to another colony is mostly owing to noise or undetected
influences of social context [28].
The model for nestmate recognition presented here
is based on a process in which individuals recognize
non-nestmates based on previous experience, with the
consequence that individuals within a colony differ in recog-
nition response. Nestmate recognition by the colony is
achieved through a distributed process, so that overall, ants
from one colony distinguish those from another, but individ-
uals do not always do so. We argue that variation makes the
system more effective and accounts for results that have until
now been taken to be owing to noise.
Our model of nestmate recognition begins with the pre-
mise that individuals acquire experience about nestmate
and non-nestmate odours over time. Ants learn odour cues
through exposure [26,29–32], over the course of hours
[33,34] and even as larvae [35], so that different ants may
react differently to the same chemical profile, and an individ-
ual ant may slowly change its reaction to a given odour.
Odour recognition in ants appears to be inclusive; ants
seem to be better at recognizing the presence of a new com-
pound rather than its absence [36,37]. Here, we consider
how nestmate recognition can emerge from the distribution
of various learned odours among ants within a colony.
2. Distributed detection modelNestmate recognition in social insects is olfactory. Hydrocar-
bons present on an insect’s cuticule (CHC) are sensed by
other workers and used to discriminate between nestmates
(individuals from the same colony) and non-nestmates
(those from other colonies). We describe the relation among
chemical profiles as in [38] with reference to cuticular hydro-
carbon or CHC-space, an n-dimensional Euclidean vector
space in which each coordinate axis represents the
concentration of a given hydrocarbon (figure 1). A particular
chemical profile is a point in this multi-dimensional space.
We assume that when an ant interacts with another one,
for example when two ants engage in antennal contact,
the focal ant can evaluate the absolute quantities of various
components in the other ant’s CHC profile [39] and esti-
mate the position of the other ant’s hydrocarbon profile in
CHC-space.
The notion of CHC-space has been used in many studies
to refer to the similarity between templates or profiles. For
example, the template similarity model [40] describes recog-
nition as a function of the distance between an individual’s
internal template, corresponding to the colony-specific CHC
profile of that individual’s colony, and the cues that individ-
ual detects from other individuals. In Newey’s [38] model,
recognition depends on the position, in CHC-space, of an
encountered odour relative to a region defined by the dis-
tance between the individual’s innate odour and that of its
colony. In the D-present and U-absent model [41], workers
use distances along axes, defined by specific desirable or
undesirable cues, to accept or reject individuals. In all of
these models, the colony odour is known to be a simplifica-
tion; individuals are known to differ slightly in CHC
profile. As Van Zweden et al. [42] point out, the results of
these models depend on how the colony odour is defined.
Here we propose that each ant carries, and can update,
the description of a boundary that partitions CHC-space
into two subsets: nestmate and non-nestmate. Each ant’s
boundary is the result of its individual history of encounters
with both nestmates and non-nestmates (figure 2). Because
ants differ in experience, each ant’s boundary is different,
and no individual ant’s boundary is the same as that of the
colony as a whole. Instead, the diverse positions of this
boundary for all ants intersect to create the colony’s profile
in CHC-space. This colony template in turn determines the
colony’s response to non-nestmates (figure 3). In the figures
we represent this boundary as a line to illustrate the idea as
simply as possible, but the shape of this boundary in the
many-dimensional CHC-space could itself vary among ants
within a colony, and is influenced by the mechanisms
involved in perception.
The figures were created using the logistic regression
model described in §2a; each boundary is initialized with a
0
0.2
0.4
0.6
0.8
1.0
0.2 0.4 0.6chemical 1
chem
ical
2
0.8 1.0
colony’s profiles
Figure 2. A single ant’s linear decision boundary in CHC-space. Any odour profileson the other side of the colony’s profiles are considered by the ant to be non-nestmates.
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
3
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
random set of weights and habituated to a set of patterns
representing the colony’s profile.
We assume that each ant is consistent in its perception of
a chemical profile over the timescale under consideration,
and that the ant classifies all odours from a particular other
colony in the same way.
Let pi,j be the proportion of individuals from colony i that
recognize j as a non-nestmate chemical profile. Then
pi,i � 0, (2:1)
0 � pi,j , 1, i = j (2:2)
and P(Rij(n)) ¼ 1� (1� pi,j)n: (2:3)
Equation (2.1) states that nestmate profiles tend to elicit no
agonistic response, although in large colonies, when there
are many nestmates that never meet, it may be that occasion-
ally an ant recognizes a nestmate as a non-nestmate. This is
probably rare since it has almost never been observed.
Equation (2.2) states that each ant can recognize only a
subset of possible non-nestmate chemical profiles. Equation
(2.3) describes the probability of the recognition event R in
which an ant of colony j is recognized as a non-nestmate
by at least one out of n individuals from colony i, and
models our assumption that each ant recognizes a different
and independent subset of chemical profiles, so that the
colony’s nestmate recognition is distributed. Because individ-
uals tend to meet particular others of certain chemical
profiles, for example nestmates of the same task group,
who may tend to share chemical profiles [43], it is unlikely
that the boundaries of ants are completely independent, but
we use this assumption to simplify our model. Equation
(2.3) describes an equilibrium that occurs on the timescale
over which each ant is consistent in its perception of a chemi-
cal profile and consistently recognizes a unique subset of
non-nestmate odours.
Finally, as a corollary of equation (2.3), the probability
that a recognition event occurs between n ants of colony iand m ants of colony j is
P(Rij(n) or R ji(m)) ¼ P(Rij(n))P(R ji(m))
þ P(Rij(n))(1� P(R ji(m)))
þ (1� P(Rij(n)))P(R ji(m)),
(2:4)
where the last two terms refer to unilateral recognition.
In the following, we use aggression as a proxy for recog-
nition, since aggression implies recognition (although we
recognize that the reverse is not true). The probability of
aggression is a fraction of the probability of R
P(Aij(n, m)) ¼ a1P(Rij(n))P(R ji(m))
þ a2P(Rij(n))(1� P(R ji(m)))
þ a3(1� P(Rij(n)))P(R ji(m)),
(2:5)
where the ais establish the probability of aggression for
bilateral and unilateral recognition.
We now consider how an ant’s boundary between
nestmates and non-nestmates in CHC-space might shift
over time. Such shifts would lead to continual changes in
the colony’s response to non-nestmates.
(a) Shift in individual recognitionAn ant’s boundary between nestmates and non-nestmates in
CHC-space can be understood as the chemical combinations
for which there are sufficient neural activity to trigger percep-
tion of a non-nestmate. We suggest that boundaries shift as a
result of a process similar to the one reported in studies
such as those of Guerrieri et al. [44] and van Zweden &
d’Ettorre [9], in which response to nestmate odours is
reduced by exposure to nestmates. The boundary is estab-
lished by habituation, and continually modified by
experience, for example by repeated encounters with an
unfriendly conspecific non-nestmate.
As an example we describe how CHC profiles could be
learned using the collective action of a metaphorical set of
neurons referred to as the ‘recognizer’, modelled as a logistic
regression [45]. Each recognizer receives the information cor-
responding to m inputs or cues and has a weight associated
with each one that determines its importance. The recognizer
fires, producing an output according to: output(x) ¼1=(1þ e�wTx), where x is the input pattern vector whose com-
ponents are individual hydrocarbons, and w is a weight
vector of real numbers that express the positive or negative
role of each hydrocarbon. The output function makes rapid
and continuous transitions between 0 and 1. We model
each ant as having an internal threshold beyond which
recognition of a non-nestmate takes place:
recognize(x) ¼ Yes; if output(x) � thresholdNo; otherwise:
�
The exact value for the threshold is not important,
so without loss of generality, assuming a threshold ¼ 0.5,
the boundary that separates recognition as non-nestmate
from recognition as nestmate is when (1=(1þ e0)) ¼(1=(1þ 1)) ¼ 0:5, which occurs when wT x* ¼ 0. These are
the values for x plotted in figures 2 and 3. Each ant’s separ-
ating boundary between nestmates and non-nestmates
arises from its particular set of weights w.
Learning in this context is the process that finds suitable
values for w. Diversity among ants arises because ants
differ in weights. We assume that over successive encounters,
ants habituate to an odour if each encounter with the odour is
not associated with aggression, or if there is a positive experi-
ence such as trophallaxis or grooming [9,23,34,44,46], while
aggression or the absence of a positive experience leads to
gradual sensitization to non-nestmate odours. This assump-
tion is supported by previous studies. Knaden et al. [47]
and Van Wilgenburg et al. [48] found that in successive
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.1 0.2 0.3 0.4 0.5chemical 1
0.6 0.7 0.8 0.9 1.00
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
1.0
0.1 0.2 0.3 0.4 0.5chemical 1
(a) (b)
chem
ical
2
0.6 0.7 0.8 0.9 1.0
Figure 3. Overall colony recognition. Each line represents a boundary in CHC-space for one individual. The region of odour profiles that will be accepted by thecolony, corresponding to the region on the nestmate side of the boundary for all individuals, is located in the small open area in the lower, central part of the figure,at about coordinates [0.58, 0.10]. The figures are based on the logistic regression model and habituation algorithm of §2a. Panels (a) 100 and (b) 500 separatingboundaries.
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
4
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
encounters between non-nestmates, aggression increased.
An exception to this was reported by Nowbahari [49], who
found that ants of Cataglyphis niger reduced aggression
towards particular individual non-nestmates in successive
encounters, suggesting that an ant may habituate to the
odour of a particular non-nestmate. However, in succes-
sive encounters with different non-nestmate individuals,
aggression did not decrease.
Weights could be adjusted by a procedure akin to the
Delta rule [50] whereby small changes to the positive and
negative weights are made in the appropriate direction in
response to each encounter. In this way, a simple mechanism,
providing each ant with a crude characterization of chemical
profiles, can collectively distinguish nestmates and non-
nestmates (figure 3) for the colony as a whole. Although
ants probably use much richer neural processes to perform
this task [33,51], a simple mechanism is sufficient to model
recognition behaviour.
3. Discussion(a) Fit with existing dataIn this section, we consider how results from the literature sup-
port our model. We focus primarily on studies that track the
response of individual workers towards non-nestmates in
repeated encounters.
Experiments by Newey et al. [26] on weaver ants
(Oecophylla smaragdina L.) show that in repeated exposures,
an individual ant reacted similarly to similar chemical pro-
files, but different ants from the same colony responded
differently to the same chemical profile. A further exper-
iment, in which an individual from the recipient colony
met workers from different colonies, showed that a particular
ant reacts differently to different odours. These findings sup-
port our assumptions that workers have a boundary that
consistently separates the same nestmate and non-nestmate
odours, that not all non-nestmate odours are perceived as
such by a given individual, and that individual workers
within a colony differ in the boundary that separates
nestmate from non-nestmate chemical profiles.
Newey et al. [26] suggest that there is a set of diverse tem-
plates based on individual odours and that recognition
requires both an individual and a colony profile, with aggres-
sive response as a function of the distance between the two
[38]. Their explanation is that each ant has encoded a different
tolerance area in CHC-space defined by both the individual’s
innate odour and the colony odour. In our model, by con-
trast, the region of rejection for an individual is not
bounded by its own odour; thus, although individuals
differ in profile, an ant with a profile closer to that of the
colony may exhibit the same aggressive behaviour as an ant
whose profile is further from that of the colony. This occurred
in a study of harvester ants (Pogonomyrmex barbatus) [23];
other studies are needed to determine whether this occurs
in many ant species and under what ecological conditions.
Roulston et al. [20] investigated how aggressive behaviour
between non-nestmates depends on the number of ants in each
trial. Their experiments report the proportion of trials that have
at least 1 aggressive encounter in assays with 1 versus 1, 1
versus 25, and 5 versus 5 live ants, as well as an assay with
1 live versus 1 dead ant. In a distributed model such as ours
and Newey’s [38], different ants identify a different subset of
chemical profiles so, for assays with few ants (1 versus 1),
we expect high variability among replicates as was observed,
and the average over many replicates to approximate the
real frequency of aggression in 1 versus 1 encounters. The dis-
tributed model would also explain why a more consistent
aggression score was observed in the 5 versus 5 assays, since
it includes several 1 versus 1 interactions.
Our model predicts that the probability of aggression in
encounters between non-nestmates depends more on the
number of different non-nestmates that each ant encounters,
than on the total number of ants or total number of inter-
actions. For example, aggression is more likely when 5 ants
of colony A meet 5 ants of colony B, than when 1 ant of
colony A meets 25 ants of colony B. In the first case, there
are 5 ants that might recognize B as a non-nestmate, while
in the second case, only 1 ant may recognize B, and if it
does not recognize one B it is unlikely to recognize the other
24. We tested this quantitatively using the data from [20]
and equation (2.3), which establishes the probability that a
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
5
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
foreign ant is recognized as a function of the number of
encounters, and equation (2.5) with a1 ¼ 1 and a2 ¼ a3 ¼ 0.
We assume only two-sided recognition because there was
less aggression in the live versus dead than live versus live
assays in Roulston et al. [20], and other work [23] suggests
that aggression is more likely when both participants recog-
nize the other as non-nestmates. This produces numbers
lower than the observed experimental averages, because one-
sided aggression does sometimes occur. We approximate pi,j,,the proportion of ants of colony i that recognize the colony
profiles of colony j as that of non-nestmates in each exper-
iment, using equation (2.5) by taking the square root of the
observed proportion of aggressive encounters in the 1 versus
1 assays for each pairing. We thus take p2i,j to reflect the prob-
ability of an aggressive act between two ants. Our predictions,
along with the results of Roulston et al. [20], are shown in elec-
tronic supplementary material, table S1. Our model yields a
good approximation for the observed ordering in the fre-
quency of aggression for the 1 versus 1, 1 versus 25 and 5
versus 5 experiments.
Although there were an equal number of possible encoun-
ters in the 1 versus 25 experiments and in the 5 versus 5
assays, the level of aggression was lower (electronic sup-
plementary material, table S1), as expected, since we
assume that a particular individual classifies all odours
from the same colony in the same way: in terms of equation
(2.3), P(Ri,j(1)) ¼ 1 2 (1 2 pi,j) ¼ pi. Conversely, each of the
1 ant’s 25 adversaries has an independent chance of recogniz-
ing the single ant; using equation (2.3), P(Rj,i(25)) ¼ 1 2 (1 2
pj,i)25. Thus the probability of observing an aggressive
encounter in the 1 versus 25 assay using equation (2.5) is
P(Ai,j(1,25)) ¼ P(Ri,j(1)) P(Rj,i(25)). In the 5 versus 5 assays,
each single ant classifies all the odours from its adversary
colony in the same way, but in contrast with the 1 versus
25 assay, both colonies have many (five) individuals in
the match. Using equation (2.3), P(Rj,i(5)) ¼ P(Ri,j(5)) ¼ 1 2
(1 2 pj,i)5 so the probability of aggression is P(Ai,j(5,5)) ¼
P(Ri,j(5)) P(Rj,i(5)). Because we assume that a single ant is con-
sistent in its classification of a colony’s odours and each ant
recognizes as foreign a distinct subset of chemical profiles,
for a given value of pi,j, P(Aij(1,1)) � P(Aij(1,25)) �P(Aij(5,5)). This result is consistent with the observations
(electronic supplementary material, table S1). There were
exceptions in two trials, which our model does not account
for: in FORb-FORs and EMI-GRF.
Next we consider how the results in Van Wilgenburg et al.[48] support our model’s assumption that ants show consist-
ent differences in which chemical profiles they recognize.
Pairs of ants from different colonies were put together, and
later, one ant from each pairing, the focal ant, was placed
with another ant of the same colony as the non-nestmate
from the previous trial. About 52% of the initial assays
resulted in aggressive behaviour, and 78% of the focal ants
showed the same behaviour, aggressive or not, during their
second encounter with a different non-nestmate of the same
colony. Results did not differ when the focal ants were
presented with non-nestmates of other colonies; however,
non-nestmate CHC profiles may have converged in response
to a similar laboratory diet [44]. These results suggest that
ants of one colony differ consistently in their sensitivity to
particular non-nestmate CHC profiles.
During the second encounter, the percentages of trials
with aggressive behaviour for the initially passive and
initially aggressive ants were 29% and 89%, respectively,
leading to an overall increase in aggressive behaviour from
52% in the first round of encounters to 60% in the second
round. There are several alternative explanations for the
observed increase. If recognition of a non-nestmate odour is
random, so that an individual is equally likely to recognize
a given odour as nestmate or non-nestmate in consecutive
encounters, then the expected distribution of passive and
aggressive ants during the second round would be the
same as in the first round. According to this explanation,
the observed distributions in [48] would have to be owing
to two distinct unexplained shifts: a moderate (19%) dimin-
ution in aggression for the initially passive ants, from 48 to
29%, and a large (37%) increase in aggression for the initially
aggressive, from 52 to 89%.
Another possibility is that the passive ants remain passive
and the aggressive ants remain aggressive, leading to 0%
aggression for the initially passive ants and 100% aggression
for the initially aggressive set. Then the observed distri-
butions would have to be produced by two unexplained
shifts: a large increase in aggression for the initially passive
ants (29%) and a moderate decrease in aggression (11%) for
those that were initially aggressive. This interpretation is
consistent with the authors’ hypothesis that exposure to an
adversary, even in the absence of a fight, increases the
likelihood of subsequent aggression.
An alternative explanation provided by our model is that
the observed distributions are owing only to one shift: a mod-
erate increase in the aggression (17%) of the initially aggressive
ants. In our model, each ant is consistent, during a short time
period, in recognizing particular odours as those of non-
nestmates, and again we assume that aggression is more
likely between two non-nestmates if both individuals recog-
nize each other as non-nestmates than if only one recognizes
the other as a non-nestmate (as in [23]). When we then calcu-
late the expected distributions of passive and aggressive ants
during the second round, assuming no overall increase in
aggression (electronic supplementary material, appendix S2),
we obtain 30% and 72% of aggressive encounters for the
initially passive and initially aggressive ants, respectively, in
contrast to the observed 29% and 89%. We suggest that the
observed increase in aggression is primarily explained by a
heightened propensity for aggression, even in the absence of
bilateral recognition, for the initially aggressive ants (as
reported by Hsu et al. [52] in other taxa). In contrast to the
experiments in [49], which showed a diminution of aggression
between particular ants in successive encounters, ants in this
study [48] did not face the same adversary in the second
round. Thus, our model provides a simple explanation for
the observed results that requires only a single process. Further
experiments are needed to test this explanation.
Guerrieri et al. [44] demonstrate that the addition of
certain components to an ant’s CHC profile make it sus-
ceptible to aggression by its untreated nestmates. This is
consistent with our predictions. Treated ants were left
together to acclimate, providing opportunities for repeated
encounters, before the bioassays were performed. Aggression
by a group of untreated ants towards a treated one was
higher than that by a group of treated ants towards an
untreated one. Our model suggests that the boundaries of
the treated ants adjusted to include the added compound,
so that treated ants were likely to recognize the untreated
ants as nestmates, because the odour of the untreated ants
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
6
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
was included in the treatment mixture. However, an
untreated ant would detect as foreign the presence of an
added compound if its decision boundary were close to
that of the colony’s odours. Such asymmetry in recognition
was reported by Bos et al. [36], and the mechanism they
propose might account for the results of Guerrieri et al. [44].
In another study, Sturgis & Gordon [23] demonstrated that
the probability of aggression depends on task group: exterior
workers are significantly more aggressive than interior
workers, and recognition is more likely towards ants of
neighbouring colonies [29,53], which are often encountered
by exterior workers [18], than ants of more distant colonies.
These results suggest that aggression is related to the past
history of encounters: individuals that have previously met
non-nestmates are more likely to recognize them [47,48].
Honeybee guards outside the hive react towards
intruders that are non-nestmates. However, recognition of
non-nestmates by bees tends to be inaccurate, with a mean
rejection rate of non-nestmates of only 51% (mean based on
Johnson et al. [21]). This is consistent with our model:
individual bees vary in the boundary in CHC-space that
separates the odours of nestmates and non-nestmates and
that recognition is distributed. Other studies show that in
the short-term, individuals are consistent in which odours
they recognize as non-nestmates [12,54], and habituation to
conspecific odours has been demonstrated in several studies
[54–56] suggesting that a bee’s recognition ability changes
as a result of its interactions with other bees.
Our model may help to explain the apparently contradic-
tory results on neighbour recognition in ants. Some studies
show that ant colonies distinguish the odour of neighbouring
colonies [29,30,57]; for example, encounters with neighbouring
foragers inhibit foraging more than encounters with ants from
distant ones. This is puzzling because field and laboratory
studies show that it is unlikely for any individual ant to
meet others of the same neighbouring colony many times
[58,59]. We suggest this effect may be owing to small shifts
in the recognition boundary of ants who have met a non-
nestmate only a few times. Consider two neighbouring
colonies, A and B, and a third colony C, so distant that ants
from A and B never encounter ants of C. Every encounter
that an ant from A has with an ant from B shifts that A ant’s
boundary slightly. However, repeated encounters between A
and B do not systematically affect the relationship between
A and C, whose ants never meet. In a test such as the one in
[29], in which large numbers of ants from A can encounter
some from B, some ants from A are likely to respond to B,
because some of them have previously met an ant from B, but
none are as likely to respond to C, because none of the A ants
have ever met any C ants. Whether such recognition results in
aggression or a ‘dear enemy’ response [30,57] may depend on
the particular ecological conditions that determine the extent
of competition and costs of aggression between neighbours.
(b) Predictions and conclusionsWe propose a distributed model of nestmate recognition that
predicts the variability in ant aggressive behaviour that is
consistently observed in experimental assays. Our model dif-
fers from other models of nestmate recognition in describing
recognition as a function of the position of a detected odour
with respect to a boundary surface in CHC-space, rather than
as a function of the distance between odours (figure 2). Our
proposal integrates features of other models: it supports
Guerrieri et al.’s [44] conclusion that ants recognize non-
nestmates rather than nestmates, and generalizes Newey’s
[38] proposal by allowing arbitrary shapes for the colony
decision boundary in CHC-space that are not bound by the
innate odours of particular individuals. Our model also
extends the D-present, U-absent [41] and U-present [9,44]
mechanisms by modelling an individual as perceiving a
chemical bouquet in which some cues can be excitatory and
some inhibitory. In contrast to previous models, in our
model each ant may be sensitive to different cues, for
example by detecting odour as a linear combination of chemi-
cals containing both positively and negatively weighted cues.
Thus, our model posits a malleable decision boundary for the
colony that emerges from the aggregated experiences of each
individual. The decision boundary could arise from a variety
of mechanisms, such as the pre-filter [60,61] or neural
template [34,62] mechanisms, which may vary among species.
We suggest that each ant in a colony learns to recognize a
unique set of non-nestmate odours, so that ants are likely to
agree more in accepting nestmates than in aggression
towards non-nestmates. Our model is similar to that of
Johnson et al. [21], in that each ant has a limited recognition
capability, and recognition of non-nestmates is the result of
the collective reaction of the colony, but differs strongly
from theirs in its assumptions and thus in its predictions
(see [63] for comments on this model). In Johnson et al.’s[21] simulation, individual recognition ability is random,
while in ours, recognition ability is a consequence of prior
experience [26,29–32,44] and thus consistent in response to
specific odours at any given time. We suggest that ants
hold a neurologically encoded decision rule that takes the
form of a separating surface in CHC-space (§2a). Thus iden-
tity is not inscribed in a template commonly shared by all
ants in a nest; it is partially encoded in each of its members
and distributed among them (figure 3).
Further work is needed to test the predictions of our
model. First, it predicts that individuals that initially do not
respond aggressively towards a particular odour can develop
an aggressive response towards that odour if it is associated
with negative experience such as aggression. This has been
found in several studies, including Knaden & Wehner [47]
and Van Wilgenburg et al. [48]. More tests of this are needed.
Second, we predict that the more an individual is exposed
to non-nestmates, the greater the precision of its discrimi-
nation of non-nestmates. In general, aggressive behaviour
should be more likely, the more non-nestmates an ant has
encountered. Repeated encounters with diverse odours will
tend to position the boundary closer, on average, to the set
of odours of its nestmates, i.e. to the colony profile. The
closer an ant’s boundary to the colony profile, the broader
the range of non-nestmate chemical profiles that it will be
able to recognize, and the more accurately it will be able to
identify foreign odours. This can be tested with experiments
involving repeated encounters, such as those in Van
Wilgenburg et al. [48], to test whether, as an ant is exposed
to more other ants, it can identify as non-nestmates ants
whose chemical profile is more similar to its own because
its boundary between nestmates and non-nestmates moves
closer to its own colony’s profile (figure 3). This prediction
is supported by the widespread observation that recognition
mistakes in the form of aggression between nestmates are
extremely rare, while mistakes in the form of a lack of
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
7
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
aggression, or failure to recognize non-nestmates, are extre-
mely common. It appears that the initial response, or
default response of a young ant, is lack of aggression. A
bias towards lack of aggression towards nestmates, because
the vast majority of a worker’s encounters are with nestmates,
favours colony cohesion at the risk of allowing intrusion or
other harm from non-nestmates. Younger ants tend to work
inside the nest [64], where they are likely to meet only
nestmates. The more similar are the odours of nestmates,
the more likely they are to be on the same side of the bound-
ary. Only later, as they work outside the nest, e.g. as foragers,
are they likely to meet ants of other colonies. Thus, as some
studies show [23], ants that leave the nest, such as foragers,
should be more likely to respond aggressively to non-
nestmates than ants that work only inside the nest, such as
brood care workers.
Third, we predict that although in general, the distances
between chemical profiles in CHC-space should be correlated
with the probability of aggression [9], recognition is not a
simple function of the distance between an individual’s
innate odour and the colony’s profile. Instead it is a function
of the distance between an ant’s decision boundary and the
intruder’s expressed odour. Thus, two odours may be very
close to each other but on distinct sides of the decision
boundary, while another two may be far from each other
yet classified together, depending on where they lie relative
to the ant’s boundary.
This implies that the relationship between the probability
of recognition and the distance between profiles in CHC-
space should change as a result of individual experience, as
shown in [36,44,65]. Our model predicts, in contrast to the
two-odour model [38], that an individual could accept, as a
nestmate, another one with a profile that is actually further
than its own from the overall colony’s profiles. Studies con-
sistent with this prediction include those of Fielde [66],
with colonies formed of mixed species combined after eclo-
sion, and the recent result that in harvester ants, CHC
profiles of some task groups were more similar within task
group, across colonies, than within colonies [23]. Further
studies such as those of [46] that measure individual odour
in isolated ants are needed to investigate this.
Evidence against our model would come from studies
that establish a lack of variation among individuals in
response to odours. An example would be the observation
that every ant in a colony identifies the same set of non-
nestmates, and that they vary only in their amount of aggres-
sion. This could be addressed using a neurophysiological
measure, rather than aggression, to determine whether recog-
nition occurs. Other evidence against our model would be the
demonstration that aggression is random with respect to
odour, without any effect of previous exposure. However,
there seems to be substantial evidence that this is not the case.
Using a distributed model of nestmate recognition
allows colonies to take advantage of large numbers of indi-
viduals to devote fewer resources to the identification of
non-nestmates. Colony hydrocarbon profiles change over
time [67,68]. A distributed system facilitates tracking of
these changes, because only a subset of ants must respond
to the change to provide an adequate response. The apparent
variability in experimental results in studies of recognition by
ants reflects not noise, but an effective method for maintain-
ing colony security. As in distributed systems elsewhere in
nature, such as in the immune system, distributed detection
allows colonies to adapt to slow changes in colony odours,
allows ants to share the energetic burden of detection and
provides the colony with a robust security system. The
system favours lack of aggression towards nestmates, risking
possible errors in the recognition of non-nestmates, in the
same way that the immune system favours the inhibition of
autoimmune responses while risking infection. A distributed
process for recognition may be important in providing group
identity to aggregates in many natural systems.
Acknowledgements. We thank Michael J. Greene, Mark Brown, Nick Bosand anonymous reviewers for very helpful comments on themanuscript.
Funding statement. We gratefully acknowledge the support provided byAsociacion Mexicana de Cultura, A.C. to F.E. and by the CiscoResearch Center University Funding program to D.M.G.
Author contributions. Both authors developed the model and wrote themanuscript. Both authors gave final approval for publication.
Conflict of interests. The authors have no competing interests.
References
1. Navlakha S, Bar-Joseph Z. 2014 Distributedinformation processing in biological andcomputational systems. Commun. ACM. 58,94 – 102. (doi:10.1145/2678280)
2. Afek Y, Alon N, Barad O, Hornstein E, Barkai N,Bar-Joseph Z. 2011 A biological solution to afundamental distributed computing problem.Science 331, 183 – 185. (doi:10.1126/science.1193210)
3. Von Boehmer H, Kisielow P. 1990 Self-nonselfdiscrimination by T cells. Science 248, 1369 – 1373.(doi:10.1126/science.1972594)
4. Cohen IR, Segel LA (eds). 2001 Design principles forthe immune system and other distributedautonomous systems (Santa Fe Institute Studies inthe Sciences of Complexity Proceedings). Oxford,UK: Oxford University Press.
5. Esponda F, Guerrero VM. 2009 Surveys withnegative questions for sensitive items. Stat. Probab.Lett. 79, 2456 – 2461. (doi:10.1016/j.spl.2009.08.019)
6. Esponda F, Forrest S, Helman P. 2004 A formalframework for positive and negative detectionschemes. Syst. Man Cybern. B IEEE Trans. 34,357 – 373. (doi:10.1109/TSMCB.2003.817026)
7. Hofmeyr SA, Forrest S. 2000 Architecture for anartificial immune system. Evol. Comput. 8,443 – 473. (doi:10.1162/106365600568257)
8. Gordon DM. 2014 The ecology of collectivebehavior. PLoS Biol. 12, e1001805. (doi:10.1371/journal.pbio.1001805)
9. Van Zweden JS, D’Ettorre P. 2010 Nestmaterecognition in social insects and the role ofhydrocarbons. Insect Hydrocarbons: Biol. Biochem.
Chem. Ecol. 11, 222 – 243. (doi:10.1017/CBO9780511711909.012)
10. Gamboa GJ, Reeve HK, Pfennig DW. 1986 Theevolution and ontogeny of nestmate recognition insocial wasps. Annu. Rev. Entomol. 31, 431 – 454.(doi:10.1146/annurev.en.31.010186.002243)
11. Dani FR, Jones GR, Destri S, Spencer SH, Turillazzi S.2001 Deciphering the recognition signaturewithin the cuticular chemical profile of paper wasps.Anim. Behav. 62, 165 – 171. (doi:10.1006/anbe.2001.1714)
12. Couvillon MJ, Ratnieks FLW. 2008 Odour transfer instingless bee marmelada (Frieseomelitta varia)demonstrates that entrance guards use an‘undesirable – absent’ recognition system. Behav.Ecol. Sociobiol. 62, 1099 – 1105. (doi:10.1007/s00265-007-0537-5)
rspb.royalsocietypublishing.orgProc.R.Soc.B
282:20142838
8
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
13. Nunes TM, Nascimento FS, Turatti IC, Lopes NP,Zucchi R. 2008 Nestmate recognition in a stinglessbee: does the similarity of chemical cues determineguard acceptance? Anim. Behav. 75, 1165 – 1171.(doi:10.1016/j.anbehav.2007.08.028)
14. Breed MD. 1983 Nestmate recognition in honeybees. Anim. Behav. 31, 86 – 91. (doi:10.1016/S0003-3472(83)80176-6)
15. Kalmus H, Ribbands C. 1952 The origin of theodours by which honeybees distinguish theircompanions. Proc. R. Soc. Lond. B 140, 50 – 59.(doi:10.1098/rspb.1952.0043)
16. Stuble KL, Rodriguez-Cabal MA, McCormick GL, JuricI, Dunn RR, Sanders NJ. 2013 Tradeoffs,competition, and coexistence in eastern deciduousforest ant communities. Oecologia 171, 981 – 992.(doi:10.1007/s00442-012-2459-9)
17. Gordon DM. 1992 How colony growth affectsforager intrusion between neighboring harvester antcolonies. Behav. Ecol. Sociobiol. 31, 417 – 427.(doi:10.1007/BF00170609)
18. Gordon DM, Kulig AW. 1996 Founding, foraging,and fighting: colony size and the spatial distributionof harvester ant nests. Ecology 77, 2393 – 2409.(doi:10.2307/2265741)
19. Van Wilgenburg E, Ryan D, Morrison P, Marriott P,Elgar M. 2006 Nest- and colony-mate recognition inpolydomous colonies of meat ants (Iridomyrmexpurpureus). Naturwissenschaften 93, 309 – 314.(doi:10.1007/s00114-006-0109-y)
20. Roulston T, Buczkowski G, Silverman J. 2003Nestmate discrimination in ants: effect of bioassayon aggressive behavior. Insectes Sociaux 50,151 – 159. (doi:10.1007/s00040-003-0624-1)
21. Johnson BR, Van Wilgenburg E, Tsutsui ND. 2011Nestmate recognition in social insects: overcomingphysiological constraints with collective decisionmaking. Behav. Ecol. Sociobiol. 65, 935 – 944.(doi:10.1007/s00265-010-1094-x)
22. Wilson E, Holldobler B. 1985 Caste-specifictechniques of defense in the polymorphic antPheidole embolopyx (Hymenoptera: Formicidae).Insectes Sociaux 32, 3 – 22. (doi:10.1007/BF02233223)
23. Sturgis SJ, Gordon DM. 2013 Aggression is taskdependent in the red harvester ant Pogonomyrmexbarbatus. Behav. Ecol. 24, 532 – 539. (doi:10.1093/beheco/ars194)
24. Van Wilgenburg E, Elgar MA. 2013 Confirmation biasin studies of nestmate recognition: a cautionary notefor research into the behaviour of animals. PLoS ONE 8,e53548. (doi:10.1371/journal.pone.0053548)
25. Larsen J, Fouks B, Bos N, D’Ettorre P, Nehring V.2014 Variation in nestmate recognition abilityamong polymorphic leaf-cutting ant workers.J. Insect Physiol. 70, 59 – 66. (doi:10.1016/j.jinsphys.2014.09.002)
26. Newey PS, Robson SK, Crozier RH. 2010 Know thineenemy: why some weaver ants do but others donot. Behav. Ecol. 21, 381 – 386. (doi:10.1093/beheco/arp201)
27. Gill KP, Van Wilgenburg E, Macmillan DL, Elgar MA.2013 Density of antennal sensilla influences efficacy
of communication in a social insect. Am. Nat. 182,834 – 840. (doi:10.1086/673712)
28. Buczkowski G, Silverman J. 2005 Context-dependentnestmate discrimination and the effect of actionthresholds on exogenous cue recognition in theArgentine ant. Anim. Behav. 69, 741 – 749. (doi:10.1016/j.anbehav.2004.06.027)
29. Gordon DM. 1989 Ants distinguish neighbors fromstrangers. Oecologia 81, 198 – 200. (doi:10.1007/BF00379806)
30. Langen TA, Tripet F, Nonacs P. 2000 The red and theblack: habituation and the dear-enemy phenomenonin two desert Pheidole ants. Behav. Ecol. Sociobiol. 48,285 – 292. (doi:10.1007/s002650000223)
31. Stroeymeyt N, Guerrieri FJ, Van Zweden JS, D’EttorreP. 2010 Rapid decision-making with side-specificperceptual discrimination in ants. PLoS ONE 5,e12377. (doi:10.1371/journal.pone.0012377)
32. Gill KP, Van Wilgenburg E, Taylor P, Elgar MA. 2012Collective retention and transmission of chemicalsignals in a social insect. Naturwissenschaften 99,245 – 248. (doi:10.1007/s00114-012-0891-7)
33. Brandstaetter AS, Rossler W, Kleineidam CJ. 2011Friends and foes from an ant brain’s point ofview—neuronal correlates of colony odors in asocial insect. PLoS ONE 6, e21383. (doi:10.1371/journal.pone.0021383)
34. Leonhardt SD, Brandstaetter AS, Kleineidam CJ.2007 Reformation process of the neuronal templatefor nestmate-recognition cues in the carpenter antCamponotus floridanus. J. Comp. Physiol. A 193,993 – 1000. (doi:10.1007/s00359-007-0252-8)
35. Signorotti L, Jaisson P, D’Ettorre P. 2014 Larvalmemory affects adult nest-mate recognition in theant Aphaenogaster senilis. Proc. R. Soc. B 281,20132579. (doi:10.1098/rspb.2013.2579)
36. Bos N, D’Ettorre P, Guerrieri FJ. 2013 Chemicalstructure of odorants and perceptual similarity inants. J. Exp. Biol. 216, 3314 – 3320. (doi:10.1242/jeb.087007)
37. Greene MJ, Gordon DM. 2007 Structural complexityof chemical recognition cues affects the perceptionof group membership in the ants Linephithemahumile and Aphaenogaster cockerelli. J. Exp. Biol.210, 897 – 905. (doi:10.1242/jeb.02706)
38. Newey P. 2011 Not one odour but two: a newmodel for nestmate recognition. J. Theor. Biol. 270,7 – 12. (doi:10.1016/j.jtbi.2010.10.029)
39. Blomquist GJ, Bagneres A-G, others. 2010 Insecthydrocarbons: biology, biochemistry, and chemicalecology. Cambridge, UK: Cambridge University Press.
40. Obin MS, Meer RKV. 1989 Mechanism of template-label matching in fire ant, Solenopsis invicta buren,nestmate recognition. Anim. Behav. 38, 430 – 435.(doi:10.1016/S0003-3472(89)80036-3)
41. Sherman PW, Reeve HK, Pfennig DW. 1997Recognition systems. In Behavioural ecology, 4thedn (eds JR Krebs, NB Davies), pp. 69–96. Oxford,UK: Blackwell Scientific.
42. van Zweden JS, Pontieri L, Pedersen JS. 2014 Astatistical approach to identify candidate cues fornestmate recognition. Front. Ecol. Evol. 2, 73.(doi:10.3389/fevo.2014.00073)
43. Wagner D et al. 1998 Task-related differences in thecuticular hydrocarbon composition of harvester ants,Pogonomyrmex barbatus. J. Chem. Ecol. 24,2021 – 2037. (doi:10.1023/A:1020781508889)
44. Guerrieri FJ, Nehring V, Jorgensen CG, Nielsen J,Galizia CG, D’Ettorre P. 2009 Ants recognize foes andnot friends. Proc. R. Soc. B 276, 2461 – 2468.(doi:10.1098/rspb.2008.1860)
45. Bishop CM, others. 2006 Pattern recognition andmachine learning. New York, NY: Springer.
46. Boulay R, Hefetz A, Soroker V, Lenoir A. 2000Camponotus fellah colony integration: workerindividuality necessitates frequent hydrocarbonexchanges. Anim. Behav. 59, 1127 – 1133. (doi:10.1006/anbe.2000.1408)
47. Knaden M, Wehner R. 2003 Nest defense andconspecific enemy recognition in the desert antCataglyphis fortis. J. Insect Behav. 16, 717 – 730.(doi:10.1023/B:JOIR.0000007706.38674.73)
48. Van Wilgenburg E, Clemencet J, Tsutsui ND. 2010Experience influences aggressive behaviour in theArgentine ant. Biol. Lett. 6, 152 – 155. (doi:10.1098/rsbl.2009.0616)
49. Nowbahari E. 2007 Learning of colonial odor in theant Cataglyphis niger (Hymenoptera; Formicidae).Learn. Behav. 35, 87 – 94. (doi:10.3758/BF03193043)
50. Rumelhart DE, McClelland JL and CORPORATE PDPResearch Group (eds). 1986 Parallel distributedprocessing: explorations in the microstructure ofcognition, vol. 1: Foundations. Cambridge, MA: MITPress.
51. Ichinose K, Lenoir A. 2010 Hydrocarbons detectionlevels in ants. Insectes Sociaux 57, 453 – 455.(doi:10.1007/s00040-010-0103-4)
52. Hsu Y, Earley RL, Wolf LL. 2006 Modulation ofaggressive behaviour by fighting experience:mechanisms and contest outcomes. Biol. Rev. 81,33 – 74. (doi:10.1017/S146479310500686X)
53. Gordon DM, Kulig A. 1998 The effect of neighbourson the mortality of harvester ant colonies. J. Anim.Ecol. 67, 141 – 148. (doi:10.1046/j.1365-2656.1998.00169.x)
54. Breed MD, Diaz PH, Lucero KD. 2004 Olfactoryinformation processing in honeybee, Apis mellifera,nestmate recognition. Anim. Behav. 68, 921 – 928.(doi:10.1016/j.anbehav.2003.10.033)
55. Dani FR, Jones GR, Corsi S, Beard R, Pradella D,Turillazzi S. 2005 Nestmate recognition cues in thehoney bee: differential importance of cuticularalkanes and alkenes. Chem. Senses 30, 477 – 489.(doi:10.1093/chemse/bji040)
56. Couvillon MJ et al. 2007 Nest-mate recognitiontemplate of guard honeybees (Apis mellifera) ismodified by wax comb transfer. Biol. Lett. 3,228 – 230. (doi:10.1098/rsbl.2006.0612)
57. Dimarco RD, Farji-Brener AG, Premoli AC. 2010 Dearenemy phenomenon in the leaf-cutting antAcromyrmex lobicornis: behavioral and geneticevidence. Behav. Ecol. 21, 304 – 310. (doi:10.1093/beheco/arp190)
58. Brown MJF, Gordon DM. 1997 Individualspecialisation and encounters between harvester ant
rspb.royalsocietypublishing.orgProc.R.Soc.
9
on April 3, 2015http://rspb.royalsocietypublishing.org/Downloaded from
colonies. Behaviour 1997, 849 – 866. (doi:10.1163/156853997X00188)
59. Adler FR, Gordon DM. 2003 Optimization, conflict,and nonoverlapping foraging ranges in ants. Am.Nat. 162, 529 – 543. (doi:10.1086/378856)
60. Ozaki M et al. 2005 Ant nestmate and non-nestmate discrimination by a chemosensorysensillum. Science 309, 311 – 314. (doi:10.1126/science.1105244)
61. Ozaki M, Hefetz A. 2014 Neural mechanismsand information processing in recognitionsystems. Insects 5, 722 – 741. (doi:10.3390/insects5040722)
62. Brandstaetter AS, Kleineidam CJ. 2011 Distributedrepresentation of social odors indicates parallel
processing in the antennal lobe of ants. J. Neurophysiol.106, 2437 – 2449. (doi:10.1152/jn.01106.2010)
63. Couvillon MJ, van Zweden JS, Ratnieks FL. 2012Model of collective decision-making in nestmaterecognition fails to account for individualdiscriminator responses and non-independentdiscriminator errors. Behav. Ecol. Sociobiol. 66,339 – 341. (doi:10.1007/s00265-011-1298-8)
64. Muscedere ML, Willey TA, Traniello JFA. 2009 Ageand task efficiency in the ant Pheidole dentata:young minor workers are not specialist nurses.Anim. Behav. 77, 911 – 918. (doi:10.1016/j.anbehav.2008.12.018)
65. Bos N, Dreier S, Jørgensen CG, Nielsen J, GuerrieriFJ, D’Ettorre P. 2012 Learning and perceptual
similarity among cuticular hydrocarbons in ants.J. Insect Physiol. 58, 138 – 146. (doi:10.1016/j.jinsphys.2011.10.010)
66. Fielde AM. 1903 Artificial mixed nests of ants. Biol.Bull. 5, 320 – 325. (doi:10.2307/1535842)
67. Provost E, Riviere G, Roux M, David Morgan E,Bagneres AG. 1993 Change in the chemicalsignature of the ant Leptothorax lichtensteinibondroit with time. Insect Biochem. Mol. Biol.23, 945 – 957. (doi:10.1016/0965-1748(93)90112-6)
68. Vander Meer RK, Saliwanchik D, Lavine B. 1989Temporal changes in colony cuticular hydrocarbonpatterns of Solenopsis invicta. J. Chem. Ecol. 15,2115 – 2125. (doi:10.1007/BF01207442)
B
2 82:20142838